2 * MPEG-4 Part 10 / AVC / H.264 HW decode acceleration through CUVID
4 * Copyright (c) 2016 Anton Khirnov
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software Foundation,
20 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
33 static void dpb_add(const H264Context *h, CUVIDH264DPBENTRY *dst, const H264Picture *src,
36 FrameDecodeData *fdd = (FrameDecodeData*)src->f->opaque_ref->data;
37 const CUVIDFrame *cf = fdd->hwaccel_priv;
39 dst->PicIdx = cf ? cf->idx : -1;
40 dst->FrameIdx = frame_idx;
41 dst->is_long_term = src->long_ref;
42 dst->not_existing = 0;
43 dst->used_for_reference = src->reference & 3;
44 dst->FieldOrderCnt[0] = src->field_poc[0];
45 dst->FieldOrderCnt[1] = src->field_poc[1];
48 static int cuvid_h264_start_frame(AVCodecContext *avctx,
49 const uint8_t *buffer, uint32_t size)
51 const H264Context *h = avctx->priv_data;
52 const PPS *pps = h->ps.pps;
53 const SPS *sps = h->ps.sps;
55 CUVIDContext *ctx = avctx->internal->hwaccel_priv_data;
56 CUVIDPICPARAMS *pp = &ctx->pic_params;
57 CUVIDH264PICPARAMS *ppc = &pp->CodecSpecific.h264;
63 ret = ff_cuvid_start_frame(avctx, h->cur_pic_ptr->f);
67 fdd = (FrameDecodeData*)h->cur_pic_ptr->f->opaque_ref->data;
68 cf = (CUVIDFrame*)fdd->hwaccel_priv;
70 *pp = (CUVIDPICPARAMS) {
71 .PicWidthInMbs = h->mb_width,
72 .FrameHeightInMbs = h->mb_height,
73 .CurrPicIdx = cf->idx,
74 .field_pic_flag = FIELD_PICTURE(h),
75 .bottom_field_flag = h->picture_structure == PICT_BOTTOM_FIELD,
76 .second_field = FIELD_PICTURE(h) && !h->first_field,
77 .ref_pic_flag = h->nal_ref_idc != 0,
80 .CodecSpecific.h264 = {
81 .log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4,
82 .pic_order_cnt_type = sps->poc_type,
83 .log2_max_pic_order_cnt_lsb_minus4 = FFMAX(sps->log2_max_poc_lsb - 4, 0),
84 .delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag,
85 .frame_mbs_only_flag = sps->frame_mbs_only_flag,
86 .direct_8x8_inference_flag = sps->direct_8x8_inference_flag,
87 .num_ref_frames = sps->ref_frame_count,
88 .residual_colour_transform_flag = sps->residual_color_transform_flag,
89 .bit_depth_luma_minus8 = sps->bit_depth_luma - 8,
90 .bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8,
91 .qpprime_y_zero_transform_bypass_flag = sps->transform_bypass,
93 .entropy_coding_mode_flag = pps->cabac,
94 .pic_order_present_flag = pps->pic_order_present,
95 .num_ref_idx_l0_active_minus1 = pps->ref_count[0] - 1,
96 .num_ref_idx_l1_active_minus1 = pps->ref_count[1] - 1,
97 .weighted_pred_flag = pps->weighted_pred,
98 .weighted_bipred_idc = pps->weighted_bipred_idc,
99 .pic_init_qp_minus26 = pps->init_qp - 26,
100 .deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present,
101 .redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present,
102 .transform_8x8_mode_flag = pps->transform_8x8_mode,
103 .MbaffFrameFlag = sps->mb_aff && !FIELD_PICTURE(h),
104 .constrained_intra_pred_flag = pps->constrained_intra_pred,
105 .chroma_qp_index_offset = pps->chroma_qp_index_offset[0],
106 .second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1],
107 .ref_pic_flag = h->nal_ref_idc != 0,
108 .frame_num = h->poc.frame_num,
109 .CurrFieldOrderCnt[0] = h->cur_pic_ptr->field_poc[0],
110 .CurrFieldOrderCnt[1] = h->cur_pic_ptr->field_poc[1],
114 memcpy(ppc->WeightScale4x4, pps->scaling_matrix4, sizeof(ppc->WeightScale4x4));
115 memcpy(ppc->WeightScale8x8[0], pps->scaling_matrix8[0], sizeof(ppc->WeightScale8x8[0]));
116 memcpy(ppc->WeightScale8x8[1], pps->scaling_matrix8[3], sizeof(ppc->WeightScale8x8[0]));
119 for (i = 0; i < h->short_ref_count; i++)
120 dpb_add(h, &ppc->dpb[dpb_size++], h->short_ref[i], h->short_ref[i]->frame_num);
121 for (i = 0; i < 16; i++) {
123 dpb_add(h, &ppc->dpb[dpb_size++], h->long_ref[i], i);
126 for (i = dpb_size; i < FF_ARRAY_ELEMS(ppc->dpb); i++)
127 ppc->dpb[i].PicIdx = -1;
132 static int cuvid_h264_decode_slice(AVCodecContext *avctx, const uint8_t *buffer,
135 CUVIDContext *ctx = avctx->internal->hwaccel_priv_data;
138 tmp = av_fast_realloc(ctx->bitstream, &ctx->bitstream_allocated,
139 ctx->bitstream_len + size + 3);
141 return AVERROR(ENOMEM);
142 ctx->bitstream = tmp;
144 tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,
145 (ctx->nb_slices + 1) * sizeof(*ctx->slice_offsets));
147 return AVERROR(ENOMEM);
148 ctx->slice_offsets = tmp;
150 AV_WB24(ctx->bitstream + ctx->bitstream_len, 1);
151 memcpy(ctx->bitstream + ctx->bitstream_len + 3, buffer, size);
152 ctx->slice_offsets[ctx->nb_slices] = ctx->bitstream_len ;
153 ctx->bitstream_len += size + 3;
159 static int cuvid_h264_decode_init(AVCodecContext *avctx)
161 const H264Context *h = avctx->priv_data;
162 const SPS *sps = h->ps.sps;
163 return ff_cuvid_decode_init(avctx, sps->ref_frame_count + sps->num_reorder_frames);
166 const AVHWAccel ff_h264_cuvid_hwaccel = {
167 .name = "h264_cuvid",
168 .type = AVMEDIA_TYPE_VIDEO,
169 .id = AV_CODEC_ID_H264,
170 .pix_fmt = AV_PIX_FMT_CUDA,
171 .start_frame = cuvid_h264_start_frame,
172 .end_frame = ff_cuvid_end_frame,
173 .decode_slice = cuvid_h264_decode_slice,
174 .init = cuvid_h264_decode_init,
175 .uninit = ff_cuvid_decode_uninit,
176 .priv_data_size = sizeof(CUVIDContext),